A poll deserves scrutiny when readers cannot identify who was surveyed, how participants were recruited, what they were asked, or when they answered.
Missing methodology does not prove a result is false. It means there is insufficient information to assess what the number represents.
Start with five questions
Who conducted and sponsored the survey? Which population was targeted? How were respondents selected? What was the exact question? When did fieldwork occur?
If the release supplies only a percentage and a logo, seek the original documentation before treating it as meaningful electoral evidence.
Self-selected votes are different
An open website or social-media poll lets interested users choose whether to participate. Its respondents can be very different from the electorate.
A large response count does not fix that mismatch. Ten thousand highly motivated visitors do not automatically represent a district better than a smaller carefully designed survey.
Pew's polling course explains why recruitment and representation matter and how weighting attempts to address imbalances. Source: Pew Research Center.
Check subgroup claims
A headline may describe the preferences of one age group even though the full survey contains relatively few people from that group.
Ask for the subgroup size and suitable uncertainty information. AAPOR's subgroup guide explains why smaller groups need particular caution. Source: AAPOR.
Sponsorship is information, not a verdict
A campaign-sponsored poll can still use documented methods. An independent label does not guarantee quality.
Report the sponsor, inspect the wording, and look for selective publication. The question is whether the evidence can be examined and compared fairly.
A practical decision rule
Use transparent surveys as evidence with limitations. Treat incomplete releases as claims requiring further information. Describe self-selected polls as responses from participants, not estimates of the entire voting population.
If a result looks surprising, neither accept it because it is dramatic nor reject it because it contradicts expectations. Check the design first.
Related reading: reading midterm polls and likely-voter models.